- Global Benefits Vision - https://www.global-benefits-vision.com/ -

AM Best warns that medical costs could outweigh health insurers’ AI savings

In November 2026, AM Best warned that US health insurers could reduce administrative work with artificial intelligence yet see little improvement in profitability if medical costs, utilisation or other expenses rose. The rating agency announced its special report on 17 November. It identifies potential gains in administration and clinical data analysis, while pointing to legacy systems, fragmented data and integration expense as barriers. It does not quantify net savings that a carrier or employee benefits plan can assume.

AM Best also says that, among 104 rated US health insurers in its innovation assessments, most were in the moderate or minimal categories and none was assessed as a leader. These categories are broader than an AI adoption survey. They do not tell an employer which insurer uses a given tool or whether an individual project has improved profitability. The report’s practical contribution is to place technology inside the operating and financial model of a health insurer.

Separate underwriting performance from operating efficiencies

An insurer’s budget should show what is expected to change in claims, administrative expense and revenue. In a claims projection, distinguish the price of care, utilisation, enrolment mix and changes to benefits. In administration, show the work eliminated, the work moved to another team and the controls required to run a new process. Estimate implementation costs separately from recurring licence fees and ongoing oversight. That makes it possible to see whether an improvement survives the full cost of introducing and operating a tool.

For example, shortening the handling time for a claim enquiry can improve service. Whether it lowers expense depends on how staff capacity is used, how much correction is needed and which costs can actually be removed or avoided. Similarly, better clinical data may help a carrier identify a trend, but a better prediction is not evidence that the insurer can reduce the underlying cost or that it should change a coverage decision. Use different measures for speed, accuracy, member experience and financial impact.

Ask what can be demonstrated at renewal

Employers should ask which gains have been measured, against what baseline and for which client population. An insurer might provide shorter response times or fewer errors for a specific workflow, but an across-the-board savings promise should show the link to fees or premiums. In a self-funded arrangement, separate administrative charges from claims that the employer bears. If claims are rising, request a clear account of what is driving them before attributing changes in spending to technology.

Also ask who reviews automated outputs that affect service or access to care. A useful assurance process names the responsible owner, defines when people must intervene and tracks corrections and complaints. Where the purchaser relies on an external administrator, establish what evidence it can receive without obtaining unnecessary patient data. Transparency about how an error is identified and corrected is at least as useful as a description of the tool.

Make the business case testable at board level

For a benefits captive that bears part of the claims risk but delegates administration, ask which entity captures any savings and which bears the costs of mistakes. Independent non-executive directors can challenge the allocation of cost and risk and ask for results split between administration and claims before approving a budget assumption. They can also request evidence from the fronting carrier or administrator where the captive cannot directly observe the workflow.

A pilot should have a stopping rule as well as a savings target. If correction work simply moves downstream or complaints become harder to resolve, the business case should reflect that deterioration. Agree what must be tracked before deployment and limit expansion until the evidence covers a representative range of claims and service requests.

The central question is not whether an insurer uses AI, but whether it can show durable improvements without weakening service or controls. A board paper should make the baseline, the measured change and the remaining uncertainty visible. That is a sounder basis for decisions on fees, renewals and underwriting assumptions than a generic productivity target.

Sources

AM Best: health insurance special report